On Sun, Jul 24, 2022 at 12:23 PM Tom Lane <tgl@sss.pgh.pa.us> wrote:
> After looking through these briefly, I'm pretty concerned about
> whether this won't break our Cygwin build in significant ways.
> For example, lorikeet reports "HAVE_SETSID 1", a condition that
> you want to replace with !WIN32. The question here is whether
> or not WIN32 is defined in a Cygwin build. I see some places
> in our code that believe it is not, but others that believe that
> it is --- and the former ones are mostly like
> #if defined(__CYGWIN__) || defined(WIN32)
> which means they wouldn't actually fail if they are wrong about that.
I spent a large chunk of today figuring out how to build PostgreSQL
under Cygwin/GCC on CI. My method for answering this question was to
put the following on the end of 192 .c files that contain the pattern
/#if.*WIN32/:
+
+#if defined(WIN32) && defined(__CYGWIN__)
+#pragma message "contradiction"
+#endif
Only one of them printed that message: dirmod.c. The reason is that
it goes out of its way to include Windows headers:
#if defined(WIN32) || defined(__CYGWIN__)
#ifndef __CYGWIN__
#include <winioctl.h>
#else
#include <windows.h>
#include <w32api/winioctl.h>
#endif
#endif
The chain <windows.h> -> <windef.h> -> <minwindef.h> leads to WIN32 here:
https://github.com/mirror/mingw-w64/blob/master/mingw-w64-headers/include/minwindef.h#L15
I'm left wondering if we should de-confuse matters by ripping out all
the checks and comments that assume that this problem is more
widespread, and then stick a big notice about it in dirmod.c, to
contain this Jekyll/Hide situation safely inside about 8 feet of
concrete.
I'll respond to your other complaints with new patches tomorrow.